이미지: METAL LAB 생성
Summary
- NVIDIA has agreed to pay $6 billion to license AI models from startup Poolside, Bloomberg reported, citing sources familiar with the matter
- The Decoder, citing an investor letter, reported that NVIDIA is extending job offers to 109 Poolside employees and licensing the company's model-building system, "Model Factory"
- Poolside's investor letter reportedly stated explicitly that the deal is neither an acquisition nor an acqui-hire, and that all three founders will stay on at the company
- 라이선스 금액
- 60억달러
- 채용 제안 직원 수
- 블룸버그 100명 이상, 더디코더 109명
- 별도 투자
- 엔비디아가 프리머니 밸류에이션 120억달러에 10억달러 투자
- 라이선스 대상
- 풀사이드가 자사 모델 제작에 쓴 '모델 팩토리' 시스템
- 관련 모델
- 직원들이 작업해온 라구나(Laguna) 모델
- 유사 선례
- 엔비디아-그록(Groq) 200억달러, 엔비디아-엔파브릭(Enfabrica) 9억달러 거래
- 자금 사용 계획
- 풀사이드는 60억달러를 내년 말까지 투자자에게 분배할 예정
- 최초 보도
- 투자자 서한을 인용한 뉴커머(Newcomer) 보도
NVIDIA has agreed to pay $6 billion to AI startup Poolside. It's not a full-blown acquisition — instead, it's a licensing deal that gives NVIDIA rights to use the AI models Poolside has built. Bloomberg first reported the news on the 20th (local time), citing people familiar with the matter, and The Decoder added further detail a day later, citing an original Newcomer report that broke the story.
What the deal actually involves
Putting together what both outlets have reported, the deal breaks down into three parts. The first is the $6 billion model license itself, which gives NVIDIA the right to use "Model Factory," the system Poolside built to train its AI models. The second is talent: according to The Decoder, 109 employees who worked on Poolside's Laguna model have received job offers from NVIDIA. Bloomberg put the number at more than 100. The third is an equity stake — separately, NVIDIA is investing $1 billion in Poolside at a pre-money valuation of $12 billion. All three co-founders are staying on to keep running the company.
Poolside's investor letter reportedly stated plainly that the deal is "neither an acquisition nor an acqui-hire," according to The Decoder. An acqui-hire is industry shorthand for a purchase aimed at people rather than the company itself. Buying a company outright typically triggers antitrust review in various jurisdictions, but by combining a license, job offers, and a minority equity stake, a buyer can secure both technology and talent without going through that process.
Why structure it this way
The Decoder noted that NVIDIA has run two similar deals before this one — a $20 billion arrangement with chip startup Groq, and a $900 million deal with networking chip company Enfabrica. All three deals share the same shape: rather than buying the company, NVIDIA licenses the technology and brings on the people.
NVIDIA makes the chips that train and run AI. But it also develops its own line of open AI models called Nemotron, and The Decoder pointed out that by acquiring Poolside's model-building system through this deal, NVIDIA could end up competing on models with the very customers who buy its chips. Poolside is reportedly planning to distribute the $6 billion it's receiving to its investors by the end of next year.
Editor's take
What stands out here isn't the price tag — it's the structure. Without buying the company, NVIDIA has carved out and separately secured the three things Poolside actually has: the system for building models, the people who know how to use it, and influence via an equity stake. Given this is the third time NVIDIA has run the same playbook, following Groq and Enfabrica, this looks less like improvisation and more like a deliberate template. A full acquisition would trigger antitrust review in the U.S. and Europe, with the risk of delay or collapse along the way. Splitting the deal into a license, job offers, and a minority stake gets to essentially the same outcome without ever facing that review.
Traditional semiconductor M&A used to mean a company's logo disappearing and its entire org chart getting absorbed. A structure like this one — where the founders stay, the company remains independent, and a chip maker simply borrows the know-how for building models — was rare just a few years ago. It's a sign that as competition for AI talent intensifies, splitting off a company's technology assets and people from the company itself, rather than buying the whole thing, is becoming the new standard.
For Korean semiconductor and AI companies, what's worth watching is that NVIDIA is separately acquiring an outside startup's model-building system even though it already has its own model line, Nemotron. A chip supplier building up its own model-making capabilities is a trend that matters for any customer that currently just buys chips — because that supplier could become a competitor. Domestic companies working across both AI chips and models would do well to study this kind of vertical integration. The next thing to watch is whether U.S. regulators take issue with this deal structure in the coming weeks, or whether it passes quietly, as the Groq and Enfabrica deals did.
Comments